ELEVATE: Agentic Controller Acquisition from Task Failure for Quadrupedal Manipulation
ELEVATE: Agentic Controller Acquisition from Task Failure for Quadrupedal Manipulation
ELEVATE is a framework that uses persistent task failures to identify and acquire missing whole-body control capabilities for quadrupedal manipulators. Starting from a single pretrained controller, it autonomously learns controllers for wall-supported elevation, supported manipulation, and wall descent across seven simulation tasks. These controllers achieve 98.4%-100% success rates in tasks beyond the initial controller's workspace and transfer to a physical robot with 18/20 successes in rack retrieval and 16/20 in floor-to-rack placement.
This paper introduces ELEVATE, a framework that enables quadrupedal manipulators to autonomously acquire new whole-body control capabilities through task failure analysis and reinforcement learning. The method demonstrates high success rates in both simulation and real-world transfer, without requiring real-robot training.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- Environment
- SimulationOpen source S4
- Control
- Reported autonomousOpen source S3
Reported numbers
Reported success rate
98.4%
View original evidence
vation, supported manipulation, and wall descent across a sequence of seven simulation tasks. These controllers are reused in subsequent reaching and object-manipulation tasks. On the five tasks beyond the initial controller's workspace, ELEVATE achieves mean success rates of 98.4%-100% across five independent runs, while coding-agent baselines with a fixed controller library fail to complete these tasks. The synthes
Open source S4Reported success rate
100%
View original evidence
vation, supported manipulation, and wall descent across a sequence of seven simulation tasks. These controllers are reused in subsequent reaching and object-manipulation tasks. On the five tasks beyond the initial controller's workspace, ELEVATE achieves mean success rates of 98.4%-100% across five independent runs, while coding-agent baselines with a fixed controller library fail to complete these tasks. The synthes
Open source S4
- paper: not_reported
- code: not_reported
- weights: not_reported
- dataset: not_reported
What remains unknown
Not established in the collected evidence: Data origin.
Reported performance applies to the described task. It does not establish general autonomy or deployment readiness.
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
vation, supported manipulation, and wall descent across a sequence of seven simulation tasks. These controllers are reused in subsequent reaching and object-manipulation tasks. On the five tasks beyond the initial controller's workspace, ELEVATE achieves mean success rates of 98.4%-100% across five independent runs, while coding-agent baselines with a fixed controller library fail to complete these tasks. The synthes
Open source S4
al capability, a coding agent formulates and iteratively refines a reinforcement-learning problem, specifying behavior-specific rewards, curricula, and termination conditions. Validated controllers are added to the capability library and composed with existing controllers to retry the task. Starting from a single pretrained loco-manipulation controller, ELEVATE autonomously acquires controllers for wall-supported ele
Open source S3
ized task programs, together with the acquired controllers, also transfer to a physical quadrupedal manipulator without real-robot training or fine-tuning, achieving 18/20 successes in rack retrieval and 16/20 in floor-to-rack placement. Videos and demonstrations are available at https://agentic-quad-elevate.github.io/public/.
Open source S5
Source：arXiv Robotics — research abstracts · arxiv.org